Below I have divided ongoing and previous work into research themes to which I was either one of the major contributors or the major contributor. Copies of the cited papers can be found in my publications' list. Please note that references starting with a C, J or B (for example [C3]) correspond to conference publications, journal publications and book chapters, respectively.
Packet forwarding techniques for relay networks (2008 to date)
Joint work with A. Tarable, Politecnico di Torino, Italy
In [C17] we introduced an accurate semi-analytical method, based on Markov chains, to compute the average erasure probability of packets in relay networks under block fading. We considered a network in which relays decide independently of each other to either receive a packet from the source (with probability p) and store it in their finite memory or randomly select a packet from their memory and forward it to the destination (with probability 1-p). We presented a detailed analysis of the proposed method and we compared simulation to theoretical results. After validating our technique, we demonstrated that the average erasure probability can be minimized by optimizing the various network parameters, such as the number of relays, the memory size at each relay and the probability p. Recent results have confirmed that the randomly select and forward approach yields a lower erasure probability than conventional first-in first-out forwarding.
Node cooperation: Performance evaluation and power control (2007 to date)
Joint work with W. Guo and I. J. Wassell, University of Cambridge, UK
At the beginning of our investigation, we considered three-node wireless networks and we performed an error rate comparison between three collaborative protocols, namely coded amplify-and-forward, coded cooperation and distributed turbo coding [C12]. In order to evaluate the performance of collaborative networks on quasi-static fading channels, we used outage probability expressions in which we set the signal-to-noise ratio (SNR) threshold equal to a value that accurately characterizes the packet error probability of the adopted transmission scheme. Theoretical expressions for the SNR threshold have been presented in [C13, J4]. Using the aforementioned approach, we assessed the error probability of symmetric networks, that is, networks with statistically similar uplink channels. Our findings for coordinated (selfish) and uncoordinated (unselfish) cooperation have been reported in [C15] and [C19], respectively.
With the aid of error rate expressions, we explored the performance of adaptive power control for cooperative networks. Initially, we considered block-coded symmetric networks [C14] and we obtained approximate expressions for the power allocation factors that minimize the bit error probability. We then extended our research to asymmetric networks using partner selection and proposed a power allocation strategy that minimizes the packet error probability. The performance gains of our proposed power control scheme over equal power allocation have been presented in [C18].
Performance-complexity tradeoff of convolutional decoding in fixed wireless access systems (2006-2007)
Joint work with A. Demosthenous, University College London, UK, M. R. D. Rodrigues, University of Porto, Portugal and I. J. Wassell, University of Cambridge, UK
In [J7] we considered the effects of amplitude quantization and finite storage space in practical Viterbi decoding implementations and we presented results showing that in fixed wireless access systems with limited antenna diversity, low memory (or, equivalently, low constraint length) convolutional codes achieve a better error-rate performance compared to that of high memory convolutional codes. Only in systems with considerable antenna diversity, can the performance of a convolutional code be improved by increasing its memory size. Nevertheless, we demonstrated that the coding advantage offered by the high memory codes is not large enough to justify the significant increase in implementation complexity. In particular, memory-2 convolutional codes achieve a coding gain of up to 1.2 dB over their memory-8 counterparts in single-input single-output fixed wireless access systems. The situation is reversed when multiple antennas are used, but the decoder of memory-8 codes occupies at least 130 times more silicon area than that of memory-2 codes.
Throughput improvement techniques for broadband fixed wireless access systems (2003-2006)
Joint work with M. R. D. Rodrigues, University of Porto, Portugal, P. Xiao, Queen's University Belfast, UK, I. J. Wassell, University of Cambridge, UK and R. Carrasco, Newcastle University, UK
We developed analytical techniques to assess the performance of broadband fixed wireless access (FWA) systems over single-input single-output (SISO) and multiple-input multiple-output (MIMO) channels. A variety of transmission technologies appropriate for broadband FWA systems have been considered such as orthogonal frequency diversity multiplexing (OFDM), single carrier-frequency domain equalization (SC-FDE), turbo coding, space-time block coding and iterative detection techniques. For example, we investigated SC-FDE schemes in [C1] and we compared their performance to that of OFDM based approaches; we concluded that SC-FDE is more efficient than OFDM only when the code rate of the error-correcting scheme approaches unity. It was also established that turbo codes offer no performance advantage over convolutional codes when used in FWA systems without antenna diversity or with limited antenna diversity. Indeed, turbo codes only outperform convolutional codes in FWA systems having significant antenna diversity [C2, J1]. We also considered time-domain turbo equalization and demonstrated that it can outperform frequency-domain solutions, such as OFDM and SC-FDE, with only a reasonable increase in the complexity due to the iterative decoding process [C11, B1].
Analysis and design of non-punctured and punctured systematic turbo codes (2002-2006)
Ph.D. research topic
The performance of a turbo code can be evaluated using the union bound on its bit error probability, a process that requires knowledge of the transfer function of the turbo code. In [C5] we introduced the concept of the augmented state diagram upon which we built a novel framework for the derivation of the full transfer function of systematic parallel turbo codes. We then extended our framework to allow for puncturing of the output bits of rate-1/3 turbo codes and we proposed puncturing patterns [C6] and interleaving techniques [C4] that can improve the error rate performance of iterative decoding at high signal-to-noise ratio values (error-floor region). A detailed description of our theoretical investigation and a thorough comparison of analytical to simulation results have been presented in [J5].
Computation of the exact union bound becomes intensive as the length of the interleaver, which separates the two constituent codes of the turbo encoder, increases. Although the complexity of the process can be reduced, simplification is not always possible in the case of punctured turbo codes. Using union bound expressions [C7] and approximations to the exact union bound which are accurate when long interleavers are considered [C8, J6], we established that puncturing patterns which can both improve the bandwidth efficiency and lower the error floor of a turbo code, can be found by means of an exhaustive search. Furthermore, we showed in [C9, J6] that a particular puncturing scheme, which we refer to as pseudo-random puncturing, can be obtained based on the generator polynomials of the turbo encoder, thereby eliminating the need for an exhaustive search among all possible puncturing patterns. Most importantly, pseudo-random puncturing ensures that the generated rate-1/2 turbo code will always yield a lower error floor than that of its parent rate-1/3 turbo code.
Last but not least, we considered the effective free distance which is a metric associated with the error-floor performance of turbo codes, and we obtained closed-form expressions for rate-1/n punctured turbo codes, both systematic and non-systematic, which we presented in [J2].